development and application of the modelling system j2000 ... · j2000 (krause, 2001, 2002), which...

Adv. Geosci., 11, 123–130, 2007www.adv-geosci.net/11/123/2007/© Author(s) 2007. This work is licensedunder a Creative Commons License.

Advances inGeosciences

Development and application of the modelling system J2000-S forthe EU-water framework directive

M. Fink 1, P. Krause1, S. Kralisch1, U. Bende-Michl2, and W.-A. Flugel1

1Institute for Geography, Jena, Germany2CSIRO Land and Water, Canberra, Australia

Received: 16 January 2007 – Revised: 3 April 2007 – Accepted: 22 May 2007 – Published: 21 June 2007

Abstract. The scientific sound definition of measures toachieve the goals of the EU water framework directive(WFD) acquires spatially distributed analyses of the wa-ter and substance dynamics in meso- to macro-scale catch-ments. For this purpose, modelling tools or systems areneeded which are robust and fast enough to be applied onsuch scales, but which are also able to simulate the impact ofchanges on single fields or small areas of a specific land usein the catchment.

To face these challenges, we combined the fully-distributed hydrological model J2000 with the nitrogen trans-port routines of the Soil Water Assessment Tool SWATmodel, which are normally applied in a semi-distributive ap-proach. With this combination, we could extend the quan-titative focus of J2000 with qualitative processes and couldovercome the semi-distributed limitation of SWAT. For theimplementation and combination of the components, weused the Jena Adaptable Modelling System JAMS (Kralischand Krause, 2006) which helped tremendously in the rela-tively rapid and easy development of the new resultant modelJ2000-S (J2000-Substance).

The modelling system was applied in the upper Gera wa-tershed, located in Thuringia, Germany. The catchment hasan area of 844 km2 and includes three of the typical land-scape forms of Thuringia. The application showed, that thenew modelling system was able to reproduce the daily hy-drological as well as the nitrogen dynamics with a sufficientquality. The paper will describe the results of the new modeland compare them with the results obtained with the originalsemi-distributed application of SWAT.

Correspondence to:M. Fink([email protected])

1 Introduction

With the commencement of the “Common ImplementationStrategy” of the European Framework Directive (EU-WFD)in the year of 2000 an European wide law is implementedto protect the various types of water bodies (Directive,2000/60/EC). One of the main objectives of the EU-WFD isto reach a good hydrochemical standard within the timescaleof 15 years. In many cases, the attainment of this good hydro-chemical target is jeopardized by adverse impacts of nutrientloads from diffuse sources.

For the analysis, prediction and management of water-sheds with nutrient problems, simulation models that de-scribe the water and nutrient dynamics might be consideredas useful tools. One widely used model for water and nutrientdynamics in catchment scale is the Soil Water AssessmentTool (SWAT) (Arnold et al., 1998). An application of SWATin the Gera catchment showed its principle applicability inthis region. Because measurements which aim at a reduc-tion of nutrients from diffuse sources, e.g. fertilizer reduc-tion, land use change, or planting of catch crops, take placeon single agricultural fields, such fields should be representedby the distribution concept of models used for their simu-lation. This is of even greater importance when the modelshould be used for prognostic scenario simulations. Becauseof its semi-distributive character SWAT was not able to repre-sent single agricultural fields as model entities. It was judgedthat a fully distributive regionalisation approach is needed,like the topological Hydrological Response Units concept ofJ2000 (Krause, 2001, 2002), which is a polygon-based ap-proach that is able to represent any geometry that is needed.To use the well tested SWAT routines to describe the nitro-gen dynamics together with the fully distributed hydrologicalcomponents of J2000, we restructured the relevant SWAT al-gorithms into JAMS-compliant JAVA components. By thecombination of these new components with the existent hy-drological parts, the new model J2000-S was formulated.

Published by Copernicus Publications on behalf of the European Geosciences Union.

124 M. Fink et al.: Development and application of the modelling system J2000-S

Fig. 1. Landuse (September 1999) of the geographic regions of the 843 km2 large Gera river System.

The work was conducted in collaboration with the“Thuringian Environmental Agency”. We choose the Geracatchment as test catchment, because it includes many repre-sentative landscape types of Thuringia.

2 Study area

For the testing of the newly developed model the upperGera catchment located in Thuringia, Germany, was selected.With an area of 844 km2, it is a typical meso-scale catchment,which is located on the northern bound of the middle moun-tain range of the Thuringian forest and south of the city ofErfurt (Fig. 1). The main streams in the catchment are theGera River itself, which drains the central part from south tonorth, the Apfelstadt River which drains the western part andthe Wipfra River which drains the eastern part of the catch-ment.

The primary reason for the selection of this catchment fordemonstrating application of this new model was the naturalconditions, in terms of land use, soils and geology which canbe considered as representative for Thuringia. The catchmentitself features three specific landscape forms (ordered fromsouthwest to the northeast):

– The “Thuringer Wald (Thuringian Forest)” (600 to980 m a.s.l.) is underlain by impermeable granite and

porphyry rocks (Seidel, 2003). The dominant land useconsists of coniferous forest. Evapotranspiration on anannual basis is estimated to be less than 30% of the an-nual precipitation, which results in a high runoff gener-ation dominated by interflow (Table1);

– The “Ilm-Saale-Ohrdrufer-Shell-Limestone” (300–600 m a.s.l.) mainly consists of underlying limestone,dolomite and calcite shales, which locally exhibitkarstfeatures. Land use is a mixture of forest, pasture,meadows and crops. The annual precipitation isreduced by 40% compared to the mountainous part;whereas, the ET is somewhat higher, which resultsin a significant reduction of runoff generation. Thedominant runoff components are interflow and baseflowdepending on specific location;

– The “Inner-thuringian agricultural hill-land” (200–300 m a.s.l.): geology exhibits an alternating stratifica-tion of marlstones and sandstones. It is overlain byloess deposits that provide a high degree of soil fertil-ity, resulting in productive crop land use. Precipitationis again reduced and ET increased, which results in aquitelow runoff generation. Because of the soil, geologyand the topographic conditions, groundwater rechargewithin this land form is the dominant runoff generationprocess.

Adv. Geosci., 11, 123–130, 2007 www.adv-geosci.net/11/123/2007/

M. Fink et al.: Development and application of the modelling system J2000-S 125

Table 1. Regional longterm water balance of the three major geographic regions within the Gera catchment (Bende-Michl et al., 2005).

Geographic region Precipitation [mm/a] Evapotranpiration [mm/a] Discharge [mm/a] Discharge quotient [–]

Thuringian Forest 1300 400 900 0.69Ilm-Saale-Ohrdrufer-Shell-Limestone 800 460 340 0.43Innerthuringian agricultural hill-land 635 540 90 0.15

Besides the natural karst features mentioned above, itshould be noted that a couple of reservoirs are located withinthe catchment; these impoundments affect the flow condi-tions exhibited within the catchment.

3 Modelling approach

In order to provide baseline simulations of the hydrologicaland the nutrient transport dynamics, the two models J2000and SWAT initially were applied independently. The re-sults of the hydrological modeling are described in detail inKrause et al. (2006) and of the SWAT application in (Bende-Michl et al., 2006).

The Soil Water Assessment Tool (SWAT) was devel-oped for the identification and quantification of (assumedand tested) nutrient source areas by the simulation of keybio-geo-chemical processes and interactions, tracking ofnutrient-load generation from land to water by main transportpathways as well as for evaluating of potential managementscenarios in meso- or large-scale catchments, as addressed bythe EU-WFD (Arnold and Fohrer, 2005; Haverkamp et al.,2004; Hormann et al., 2005).

The J2000 is a modular process-oriented hydrological sys-tem, which implements single hydrological processes as en-capsulated process modules. Beside the process modulesfor the simulation of the runoff generation and runoff con-centration dynamics, J2K offers routines for the regionalisa-tion and correction of climate data, the calculation of addi-tional input data e.g. solar radiation, potential evapotranspi-ration according to Penman-Monteith or absolute humidity,which are often not available as measured values. In addi-tion, tools for sensitivity and uncertainty analysis as well asplotting capabilities to produce diagrams of the model re-sults are available. The temporal resolution of J2000 is ineither daily or hourly time steps (specified by the user). Spa-tial resolution or catchment distribution is made using thehydrological response units (HRUs) which are connected bya lateral routing scheme to simulate lateral water transportprocesses. This allows a fully-distributed hydrological mod-elling of catchments (Krause et al., 2006). The implementa-tion of the nitrogen-balance algorithms of SWAT in J2000-Sis described atBende-Michl et al.(2006).

Because of the geological conditions, karstic features ap-pear in part of the Gera catchment. as river ’sinks’ (reacheslosing water to the channel) which are delineated for the

different tributaries and the mainstem Gera River. Unfortu-nately the total or relative amount of water that is lost in thesesink areas has never been measured or estimated. Nonethe-less, their effects can be estimated from a water-balance anal-ysis, in particular, during low-flow periods. To account forsuch processes, we developed and implemented a conceptualapproach (1) to describe such flow losses, as:

Qd = L ∗ RH 2∗ Ksink (1)

WhereQd is the amount of water that leaves the systemdue to a channel sink,L is the length of the stream segment,RH is the hydraulic radius andKsink is a calibration param-eter which can be related to the hydraulic conductivity of thestream bed.RH is squared because when the water levelrises the stream width generally increases as well. Also, thegradient from the stream’s water surface to the groundwaterlayer increases with higher water level which conceptuallyis described by the square ofRH . The amount of nitrogenthereby removed from the streams is calculated with the helpof actual nitrogen concentrations in the affected stream seg-ment. This water and nutrient-loss aspect applies only in thelimestone regions as shown in Fig.2.

Because the spatial resolution of the semi-distributedmodel SWAT with its unlocated HRUs and subcatchments(Fig. 3), we implemented the relevant algorithms of SWATinto the fully distributive model J2000 in order to gain ahigher resolution of results. The implementation of the ni-trogen balance algorithms of SWAT in J2000-S is describedby (Bende-Michl et al., 2006).

3.1 Input data

The application of both models requires GIS-based data,such as topography, geology, soils and land-use information.A 25×25 m DEM was available to delineate the subcatch-ments, their effective slope and aspect, and soil informationwas derived based on the soil map “Die LeitbodenformenThuringens” 1:100 000 (Rau et al., 1995). From the legenddescription for the soil horizons, field capacity and hydraulicconductivity were extracted. Information of the underlyingbed rock was obtained from the geological map (“Geolo-gischeUbersichtskarte 1:200 000”), which was also used toidentify regions with karst features. Land cover informationwas derived from two land-use classifications (Fig.1) fromLandsat TM images of the years 1999 and 2002. For theinitial SWAT modelling application, information regarding

www.adv-geosci.net/11/123/2007/ Adv. Geosci., 11, 123–130, 2007


Fig. 2. Limestone regions (yellow) and rivers where deep sink is active(cyan) of the Gera catchment.

Fig. 3. Distribution concepts of SWAT and J2000.

land-use management was using classification for the year2002 only. For the J2000-S modelling application, infor-mation was obtained by a fusion of different maps from therelevant Thuringian state agencies; however, both methodsshowed very similar results. Consumption of crop specificfertilizer was estimated from literature studies (TLL, 2001)and from farmer interviews conducted in a nearby catchment(Fink, 2004). The results used as model inputs are shown inTable2.

Climatological time series data were available at three 3synoptic climate stations and 14 precipitation stations in thevicinity of the Gera catchment. Runoff data from four gauges

Table 2. Crop types and fertilization for different regions of theGera catchment (Bende-Michl et al., 2005).

Crop [kgN/ha∗ a] [kgN/ha∗ a]“intensive” “less intensive”

Winter wheat 178 164Maize 202 –Summer barley 106 –Winter barley 178 164Rape 202 176Peas – 50Fallow – 0Meadow 60 60

were used for model calibration and validation. All hydro-meteorological data were available in daily time steps. Hy-drochemical datasets of interest comprised of, besides majorcations and anions, the nitrogen components, such as ammo-nia, nitrite and nitrate. Such data were available from mea-suring network of the state Thuringia.

3.2 Modelling results SWAT

For the baseline simulation, SWAT initially was applied inits original form. Figure4 compares the predicted and mea-sured runoff for the period of 1/1994 to 11/2000. It is obvious



Fig. 4. Predicted and measured discharge at the catchment outlet(Mobisburg).

from the plot that, during low-flow periods, the simulatedrunoff showed a significant overprediction much of the time.Comparable peak flows matched better but did also exhibitoccasionally poor model results. In general, it can be con-cluded that the hydrological dynamic processes of the Geracatchment could not be represented well with SWAT. Thiscan also be seen from the long-term average runoff whichgave a simulated average of 6.8 m3/s; whereas, the measureddischarge was only 5.8 m3/s. This overprediction of 17% canto some extent be related to the effects of the karst featuresas discussed above. Such discharge losses can not be de-scribed adequately in SWAT. The efficiency “Eff” (Nash andSutcliffe, 1970) for the whole period was 0.51 which com-prises “Eff” of 0.27 for the calibration period (1998–2000)and 0.60 for the validation period (1994–1997).

The SWAT modelling results of the nitrogen-load dynam-ics are given in Fig.5. The figure indicates in general thatthe level and the dynamics of the prediction are more or lesscorresponding to the measured values but overall were over-predicted during low-flow periods. The direct correlationof the measured values with the prediction of the nitrogenload on a daily basis resulted in a poorr2 value of 0.25 forthe entire period. During the calibration period (1998–2000)r2 resulted in a value of 0.36 and during the validation pe-riod (1995–1997) of 0.28. Similar values were calculatedfor the nitrogen-concentration dynamics, which gave a coef-ficient of determination of 0.29 for the entire period. As ajustification of these low values, it has to be noted that thenumber of prediction-observation pairs used for the calcula-tion of the efficiency measures was relatively low becauseof the rather large time interval between measurements. Thecomparison of the simulated long-term average nitrate con-centration of 19.8 mg/l agreed well with the measured valueof 22.2 mg NO3/l. A part of 1.4 mg NO3/l of the predictedvalue was added from data of the nitrogen loads of wastew-ater treatment plants. The yearly load of these point sourcesaccounts for a total of 64487 kg N/a.

Fig. 5. Predicted and measured nitrogen load at the catchmentoutlet (Mobisburg).

The spatial representation of the modelled nitrogen loadis shown in in Fig.6. The figure shows that higher loadsfor the southern part of the catchment and lower loads forthe northern part were modelled. This indicates that the loaddistribution in the Gera catchment is highly dominated bythe runoff generation processes, already discussed in Sect.2,which explains the contradiction between land use and ni-trogen load production. The high values in the middle ofthe southern part of the catchment are caused by a combina-tion of agricultural land use, relatively poor soils and distinctrunoff generation. In addition the figure shows the relativepoor distribution by subbasins, which is used by SWAT, andwhich does not allow the detailed identification of responsi-ble source areas.

3.3 Modelling results J2000-S

The runoff modelled with the J2000-S (Fig.7) showed amuch better fit than the SWAT-modelling results (Fig. 3).This can also be seen by the higher Nash-Sutcliffe efficiencyof 0.71 and the lower deviation of the simulated long-termmean runoff (5.9 m3/s) from the observed one (5.8 m3/s).The good agreement in the long-term mean indictes that itwas possible to consider the flow losses by the karst phenom-ena well with in the model extension described in Sect.3.The figure indicates that the model results tend to overpre-dict single peak-flow events. This can be explained by theinfluence of the various water reservoirs in the catchmentswhich is not considered in the model.

The better representation of the hydrological dynamics,together with the higher spatial resolution of the model,resulted also in a better representation of the simulatednitrogen-load dynamics as shown in Fig.8. A coefficient ofdetermination of 0.64 could be achieved. The fact that thesimulation of the nitrogen concentration produced a valueof r2=0.28, which is very similar to the SWAT simula-tion, indicated that the major reason for the model improve-ment in terms of nitrogen load is related to the the betterrepresentation of the hydrological dynamic. Nevertheless,



Fig. 6. Predicted average annual nitrogen load leached within the derived subcatchments in the Gera catchment (Bende-Michl et al., 2005).

Fig. 7. Predicted and measured discharge at the catchment outlet(Mobisburg).

the long-term modelled average nitrate concentration of23.9 mg/l is still very close to to the observed value of22.2 mg NO3/l.

The most important difference between the results fromthe two models, which was also the driving motivation for thewhole work, is the better spatial representation of the modelresults. This is shown in Fig.9 which shows the distributedresults of the nitrogen output loads using the topographicalHRUs of J2000-S. The comparison of this map with the orig-

Fig. 8. Predicted and measured nitrogen load at the catchmentoutlet (Mobisburg).

inal SWAT results (Fig.6) showed many similarities but ina higher spatial resolution. For example, the SWAT map didshow 3(4) subbasins with high loads in the middle southernpart of the catchment. The better spatially resolved represen-tation of J2000-S could explain that those higher loads werecaused by a couple of localized areas in the catchment whereagricultural use on poor soils with a low retention capacity,occurred.



Fig. 9. Predicted average annual nitrogen load leached within the derived HRUs in the Gera catchment.

4 Summary and conclusions

The general objective of the study presented here, was toanalyse and assess the requirements needed to model diffusenutrient transport processes in a high spatial distribution inmeso-scale catchments. Such modelling is needed for thedevelopment of management plans for the EU Water Frame-work Directive.

To provide base line modelling scenarios we firstly ap-plied the hydrological model J2000 and the nutrient transportmodel SWAT in the Gera catchment in Germany. To use thehigher spatial resolution of J2000 with the nutrient transportprocesses of SWAT we extracted the necessary algorithmsand refactured them as JAMS compliant process modules.The combination of the relevant process descriptions fromboth models resulted in the J2000-S model which was thenused for the same catchment and the same time period.

The comparison of the results from both models showednot only differences in the spatial resolution but also in thequality of the simulation in terms of hydrology and nitro-gen transport. With the J2000-S model a Nash-Sutcliffeefficiency for the hydrological dynamics of 0.71 could beachieved which was very much better than those of SWATmodel (0.51). The quality of the simulation of nitrogen loadscould also be improved by the new model approach which isshown by an increase ofr2 from 0.25 to 0.64. As the simu-lation of the nitrogen concentrations did not change signifi-cantly with the new model it can be stated that the increase in

the nitrogen load simulation quality can mostly be related tothe better reproduction of the hydrological dynamics. Alas,both models had problems with the N-concentration simula-tion, but the long-term balance was met well.

The results show that J2000-S may serve as an adequatetool to support the catchment-related work that has to be donewithin the scope of the EU-WFD. In particular the higherspatial resolution can be considered as a very important pre-liminary to derive suitable management scenarios. Furtherimprovements in this new model will be the implementationof the phosphorous cycle and of stream water quality rou-tines.

Acknowledgements.The authors wish to acknowledge the kindcooperation with the “Thuringian Environmental Agency”, Jena,Germany.

Edited by: K.-E. LindenschmidtReviewed by: C. Gattke and Y. Wang

References

Arnold, J., Srinivasan, R., Muttiah, R., and Williams, J.: Large AreaHydrologic Modelling and Assessment. Part I, Model Develop-ment, J. Am. Water Resour. As., 34(1), 73–89, 1998.

Arnold, J. R. and Fohrer, N.: SWAT2000: current capabilities andresearch opportunities in applied watershed modelling, Hydrol.Processes, 19, 563–572, 2005.

Bende-Michl, U., Fink, M., and Flugel, W.-A.: Application of theSWAT-model for simulation and assessing landuse management



actions according to the European-Framework Directive (EU-FDW), in: MODSIM 2005 International Congress on Modellingand Simulation Modelling and Simulation Society of Australiaand New Zealand, edited by: Zerger, A. and Argent, R. M.,2686–2692, 2005.

Bende-Michl, U., Krause, P., Kralisch, S., Fink, M., and Flugel,W.-A.: Current development and application of the modular Javabased model JAMS to meet the targets of the EU-WFD in Ger-many, in: Proceedings of the iEMSs Third Biennial Meeting:“Summit on Environmental Modelling and Software”, edited by:Voinov, A., Jakeman, A., and Rizzoli, A., International Environ-mental Modelling and Software Society, Burlington, USA, 2006.

Directive 2000/60/EC of the European Parliament and of the Coun-cil of 23 October 2000 establishing a framework for Communityaction in the field of water policy, Official Journal of the Euro-pean Communities, L327/1, 2000.

Fink, M.: Regionale Modellierung der Wasser- und Stick-stoffdynamik als Entscheidungsunterstutzung fur die Reduk-tion des N-Eintrags am Beispiel des Trinkwassertalsperren-systems Weida-Zeulenroda, Thuringen, Jena Friedrich-Schiller-University, Chemisch-Geowissenschaftliche Fakultat, Disserta-tion, 2004.

Haverkamp, S., Fohrer, N., and Frede, H.-G.: Assessment of theeffect of land use patterns on hydrologic landscape functions:a comprehensive GIS-based tool to minimize model uncertaintyresulting from spatial aggregation, Hydrol. Processes, 19, 715–727, 2005.

Hormann, G., Horn, A., and Fohrer, N.: The evaluation of land-use options in mesoscale catchments. Prospects and limitationsof ecohydrological models, Eco. Mod. 187, 3–14, 2005.

Kralisch, S. and Krause, P.: JAMS – A Framework for Natural Re-source Model Development and Application, in: Proceedings ofthe iEMSs Third Biennial Meeting: “Summit on Environmen-tal Modelling and Software”, edited by: Voinov, A., Jakeman,A., and Rizzoli, A., International Environmental Modelling andSoftware Society, Burlington, USA, 2006.

Krause, P., Base, F., Bende-Michl, U., Fink, M., Flugel, W., andPfennig, B.: Multiscale investigations in a mesoscale catchment– hydrological modelling in the Gera catchment, Adv. Geosci.,9, 53–61, 2006,http://www.adv-geosci.net/9/53/2006/.

Krause, P.: Das hydrologische Modellsystem J2000 – Beschrei-bung und Anwendung in groen Flugebieten, Schriften desForschungszentrums Julich, Reihe Umwelt/Environment, 29,2001.

Krause, P.: Quantifying the impact of land use changes on the waterbalance of large catchments using the J2000 model, Phys. Chem.Earth, 27, 663–667, 2002.

Nash, J. and Sutcliffe, J.: River Flow Forecasting through Concep-tual Models – Part I: A Discussion of Principles, Jo. Hydrol., 10,282–290, 1970.

Rau, D., Schramm, H., and Wunderlich, J.: Die LeitbodenformenThuringens, Geowissenschaftliche Mitteilungen von Thuringen,Beiheft 3, 1995.

Seidel, G. (Ed.): Geologie von Thuringen, Stuttgart, 2003.Thuringer Landesanstalt fur Landwirtschaft (TLL): Anleitung

und Richtwerte fur Nahrstoffvergleiche nach Dungeverodnung,Thuringer Landesanstalt fur Landwirtschaft, Jena, 2001.


http://www.adv-geosci.net/9/53/2006/

development and application of the modelling system j2000 ... · j2000 (krause, 2001, 2002), which...

Documents